Hi there 👋

• 🌎 ⇌💡⇌ 💻 ⇌ 🚀
• “Code is like humor. When you have to explain it, it’s bad.” – Cory House

---

Overview

• 🧠 MM Foundation Model for Decision Making

  • AdFlux Engine: Simulates user behavior; leverages LAVA, Gato, RLHF for ad optimization.
  • AdFlux PersonaTaste Engine: Utilizes shopping data, web clicks for personalized ad experiences.
  • AdFlux Agentic Engine: Sophisticated ad system using agentic decision-making models.
  • Agentic Humanoid: Autonomous components collaborate for dynamic, adaptive decision-making.

• 🚀 MMLLM Post Training

  • SpectraQLoRA: Combines QLoRA and Spectrum for efficient fine-tuning.
  • Hybrid SpectraQLoRA: Adds dynamic adapter activation based on SNR.
  • Adaptive Low-Rank Spectrum Fine-Tuning (ALoRS): Adjusts LoRA ranks based on layer importance dynamically.
  • Spectral Gradient Merging (SGM) : Merges gradients across similar high-SNR layers.

• 🎯 Alignment

  • Model Alignment Overview
  • AAEP-RedTeaming: AAEP identifies, assesses, and addresses LLMs' adversarial vulnerabilities.
  • ASSC-Alignment: AI alignment via self-supervised Large Language Models.
  • llm-eval-framework: Evaluation Paradigm for Aligning LLM Behavior with Real-World User Expectations
  • llm-rlhf-optimization: A Scalable Framework for Human-Centric Model Adaptation

• 🧩 Reasoning

  • Reasoning Models Overview

• 🔍 Interpretability

  • Model Interpretability Approaches

• 🛡️ Responsible AI: Safety, Privacy, Trust and Ethics

  • Responsible AI Approaches
  • AAEP-RedTeaming: AAEP identifies, assesses, and addresses LLMs' adversarial vulnerabilities.
  • llm-eval-framework: Evaluation Paradigm for Aligning LLM Behavior with Real-World User Expectations
  • llm-rlhf-optimization: A Scalable Framework for Human-Centric Model Adaptation

• 🕹️ Agents and Agentic

  • Agentic Humanoid: Autonomous components collaborate for dynamic, adaptive decision-making.
  • AdFlux Agentic Engine: Sophisticated ad system using agentic decision-making models.
  • Agentic Decision Making: Autonomous AI systems making informed, independent decisions.

• 🌍 World Modeling

  • BindingWorldModel: Fuses multi-sensory data for unified intelligent representations.
  • Realator: A Real simulator for the purpose of world model building.
  • MM4WM: MM4WM unites model training and learning for enhanced physical intelligence.

• 🤖 Embodied AI

  • Agentic Humanoid: Autonomous components collaborate for dynamic, adaptive decision-making.
  • Humanoid Foundation Model: Predicting action sequences through video-based joint analysis.

• 📚 RAG (Retrieval Augmented Generation)

  • RAG Methods Overview

• ⭐ Cool

  • DiffuNet: Neural Architectures with Diffusion Models